See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/247334096

Rational Actors, Equilibrium and Social Institutions

Article · January 1995

CITATIONS READS

223 1,479

1 author:

Randall Calvert
Washington University in St. Louis
37 PUBLICATIONS   2,703 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Political Economy of Regulation View project

Rationality and identity View project

All content following this page was uploaded by Randall Calvert on 11 April 2020.

The user has requested enhancement of the downloaded file.

56 Explaining Social Institutions

30. For the imperfect monitoring approach, see the pioneering work by
Green and Porter (1984). For refinements of this approach, see Abreu, Pearce,
and Stacchetti 1986 and Abreu, Milgrom, and Pearce 1991.
31. Weiner 1932,218.
32. De Roover 1965, 114; Dollinger 1970,86.
33. Urkundenbuch der Stadt Lubeck, vol. 1, no. 156, 371, translated by
Rational Actors, Equilibrium, and Social Institutions
Dollinger 1970,383.
34. Dollinger 1970,48-51. Randall L. Calvert
35. Dollinger 1970,49; Weiner 1932,219.
36. Dollinger 1970, 51.
37. Dollinger 1970,63.
38. Dollinger 1970, 66. For further details of this embargo, see Dollinger
1970, 63-66 and Weiner 1932,220. 1. Introduction
39. For the relative size of Italian and German cities, see Bairoch, Batou, and
Chevre 1988. Some intercity cooperation was also practiced among the Italians,
where smaller cities "affiliated" themselves with larger ones. See the discussion
In their broadest, sociological definition, institutions include "all the
in the text.
beliefs and modes of conduct instituted by the collectivity" (Durkheim
40. Note that this formulation captures the gains to the ruler from either 1938 [1895]),1 that is, beliefs and conduct not depending solely on the
abusing rights directly or from neglecting to provide merchants with costly judgment, will, or habits of individuals considered in isolation. So de­
protection. fined, institutions range from the simplest informal norms to the most
41. This result is not an artifact of our specification of costs. For example, if we complex formal organizations. For Durkheim, the matter of definition
had specified that the costs borne by the city in�luded some fixed costs per trader ended there: the collectivity served as the basic unit of explanation, and
(possibly in addition to the proportional costs), the city would have an even institutions as the primitives for explaining other social phenomena.
stronger incentive to reduce the number of traders, because it bears only a This approach, however, begs two compelling questions. First, why does
fraction ( r) of the resulting Joss of value but saves all of the service costs. Making a given institution succeed in constraining individual behavior in some
costs proportional to value minimizes the distortion in the city's incentives, but
circumstances, but not in others? Second, is there any limit to the combi­
still leaves it tempted to seek short-term gains by cutting services at the expense of
nations of beliefs and conduct that can persist as institutions?
individual traders when only the bilateral reputation mechanism is at work.
The rational-actor, or rational-choice, approach to understanding so­
42. Kandori 1992.
43. This is a Nash equilibrium of the game with the properties that (1) the cial behavior is nicely suited to addressing these questions, and to studying
players's strategies at any date depend only on whether a boycott has been
the behavior of individuals in institutions. It focuses on the conflict be­
announced and (2) each player's strategy at each date maximizes his payoff tween individual motives and social prescriptions; without this conflict
from that date onward, given the equilibrium strategies of the other players. there would be no problem of institutional effectiveness or persistence.
44. In addition to the evidence mentioned in the text, see Carus-Wilson 1967, The purpose of this chapter is to examine how the rational-choice ap­
xviii and EHD 1975, 515-16. The role of the guild in securing rights rather than proach can be most effectively applied to address such questions, and to ill­
in achieving privileges in Bruges is suggested by the city policy of providing all ustrate how both informal and formal institutions can exist as equilibria in
nations with the same rights. See De Roover 1948,15. the context of repeated cooperation games. Along the way, it will be possi­
45. For the relations between the Hansa and England during this later period,
ble to suggest a more precise definition of "institutions" using the rational­
see Colvin 1971; Postan 1973.
choice model to distinguish the systems of behavior we normally think of as
46. See previous discussion of the Hansa embargoes during the late four-
teenth century. Regarding the English traders, see Lloyd 1991; Dollinger 1970.
institutions from systems that normally would not receive that title .
For a general discussion, see Greif 1992b.
Rational-Choice Approaches to Institutions
47. For details, see Bien 1989; Hoffman 1989; and Root 1989.
48. This argument is developed at length in Weingast 1992.
Within the rational-choice approach to studying institutions, three gen­
eral methods can be discerned in the recent literature. One approach

57
58 Explaining Social Institutions
views institutions as features of individual preferences: the actor's utility
function includes arguments for altruism, cooperation, or observance of
social norms, and optimal behavior under such a utility function is
worked out. Examples include the work of Margolis 1982 and the essays
in Koford and Miller 1991. Until an explanation is given for why prefer­
ences take these forms, however, this approach is really closer to the
Durkheimian one than to rational-actor methods, in that it takes the
needs of the social group as the basis for individual behavior. In princi­
ple, an explanation for such prosocial preferences in terms of basic
psychological mechanisms could bring this approach back to the level of
individual action, but a coherent psychological theory of this type has
yet to emerge.2 For the present study, I ignore this approach in favor of
more fundamentally rational-choice ones.
The other two approaches adhere more closely to the rational-actor
idea. One is to treat institutions as constraints on behavior, or "rules of
the game," and to examine rational behavior within the constraints im­
posed by institutional rules. Examples can be found in North 1981, 1990;
Shepsle 1979; and Shepsle and Weingast 1987. In this literature, the
rules themselves are either taken as given, or else treated as alternative
sets of constraints to be imposed by the creators of constitutions or of
basic procedural rules, guided by the results those rules will yield once
rational behavior takes place under them.
The third approach, originally suggested by Schotter ( 1981), re­
gards an institution as an equilibrium of behavior in an underlying game.
The aim is to describe institutions in a way that both explains the ob­
served behavior and identifies conditions under which the institution can
be effective. The present essay employs this institution-as-equilibrium
approach, and uses it to examine the differences between formal and
informal institutions, the circumstances under which they can exist, and
their effects on the welfare of participants.
The institutions-as-constraints approach is useful for understanding
behavior under stable institutions, but it falls short as a tool for under­
standing institutions themselves. Its failure lies in the fact that it takes as
given the institution's effectiveness in channeling behavior. "Rules of the
game" are either (a) impossible to violate, due to the specification of the
game or the guaranteed actions of other players; or (b) backed up by some
form of external sanction} Either form of rule simply pushes the problem
of institutional effectiveness and persistence back to another level. How is
the choice of actions constrained? Why do others behave in the way that
they do? How is the appropriate application of sanctions assured? A full
rational-choice explanation of institutions requires that the behavior of
each actor, including those applying the sanctions, be consistent with that
Rational Actors, Equilibrium, and Social Institutions 59
actor's self-interest. Moreover, any discussion of institutional choice or
institutional change in the "institutions-as-constraints" model must neces­
sarily be ad hoc.
In the institutions-as-equilibria approach, however, the game is )ust
a description of underlying physical realities: if people behave in a cer­
tain combination of ways, nature responds with certain goods or condi­
tions. Any additional structure "instituted by the collectivity" must be
described as the behavior patterns of individuals and their expectations
about the behavior of others. These patterns of behavior and expecta­
tions must be consistent with utility maximization by each individual.4 In
short, an institution is an equilibrium in the underlying game, and differ­
ent institutional structures correspond to different equilibria, while the
rules of the game remain constant. In particular, if the underlying game
(that is, nature) does not set apart any individual players as having
special opportunities or powers, then such role differentiation can be
maintained only as part of an equilibrium.
Not every equilibrium behavior pattern is an institution, however.
A game may have atomistic or anomie equilibria in which no player's
actions have any bearing on any other's actions, such as unconditional
defection by all players in a repeated many-player prisoner's dilemma.
Such a pattern of behavior is certainly not what we normally mean by an
"institution," although some authors discuss a "norm" of selfish, calculat­
ing rationality ( e.g., Marsh and Olsen 1989) tantamount to an institu­
tion. Even when the behavior of each member of the group depends on
that of a few others, the resulting pattern may not really qualify as an
institution. For example, consider a situation of repeated random match­
ing of members of a large group, in which each matched pair engages in
a two-player prisoner's dilemma, after which the players are randomly
matched again, and so on. This game has an equilibrium ( shown in
section 2 below) in which each pair of players maintains a relationship of
reciprocal cooperation, but failure to cooperate is punished by only the
partner who was cheated and not by other group members. Such a
pattern of behavior is not really an "institution" in the large group, since
players' expectations about other players' behavior do not generally
have any bearing on that behavior.s ( Reciprocity would properly be a
small-scale institution within each cooperating pair of players, however.)
On the other hand, if failure to cooperate with one ( previously coopera­
tive) partner were punished by all subsequent partners, an institution of
reciprocity would be present in the large group. The institutions-as­
equilibria model suggests that we define an institution within a group of
individuals as an equilibrium in which individuals' actions are dependent
upon the past actions of many others, or upon expectations about the
60 Explaining Social Institutions
future reactions of many others, to one's present actions.These reac­
tions could include instances of inaction or acquiescence, such as a will­
ingness not to punish a player who imposes an acceptable punishment on
a deviating third player. 6
Within the class of equilibria that qualify as institutions by this crite­
rion, there are institutions with various levels of formality and organiza­
tional complexity. A universally sanctioned reciprocity relationship, for
example, is an institution in which all members of the group use the same
strategy; the institution involves no differentiation of roles. On the other
hand, many institutions involve official positions, with rights and responsi­
bilities unique to each office.Members of a group may be sanctioned for
failing to act in accord with some official's actions, but failure of the
official to act within prescribed limits may incur sanctions as well. At the
extreme of such complexity lie those institutions that are referred to as
"organizations." In addition, institutions differ as to the formality of their
specification by the players. Institutions other than those at the highest
levels of organizational complexity may be defined by nothing more than
an informal and unvoiced understanding among the group's members, or
by a formal written contract or law, or by any level of formality in be­
tween.Finally, institutions at all levels of organizational complexity and
all levels of formality of specification may be created through a specific
process of negotiation and agreement (the U.S. Congress, for example);
or they may arise spontaneously by the accretion of practices and prece­
dents over years or centuries (the common law, for example), or by the
accidental coordination of like-minded individuals. All institutions, how­
ever, must have a common property: it must be rational for nearly every
individual to almost always adhere to the behavioral prescriptions of the
institution, given that nearly all other individuals are doing so.
Plan of the Chapter
I employ a basic idea of Milgram, North, and Weingast 1990 to portray
various possible institutions in a standard "social dilemma" situation.7
Section 2 presents a basic model of repeated cooperation built around
the two-player prisoner's dilemma (PD). In each iteration of this game,
players in a group are paired up at random to play the PD, so (unlike the
"Townsend Thrnpike" configuration used by Milgram et al.) coopera­
tion is possible without communication, using only a rule of pairwise
reciprocity in which deviation is not punished by the group at large.
Conditions are derived for a kind of tit-for-tat strategy to be in equilib­
rium.This noninstitution equilibrium then provides a point of compari­
son for the more highly organized equilibria described subsequently.
Rational Actors, Equilibrium, and Social Institutions 61
Section 3 enriches the model to provide opportunities for the play­
ers to communicate, but (again unlike the model of Milgram et a!.) such
communication is costly. Under demonstrated conditions, I show how
cooperation may be possible with such multilateral communication but
impossible under the no-communication conditions of section 2. The
equilibrium that accomplishes this feat is a true institution, in that the
whole group monitors and punishes deviant behavior. Then, in section
4, I construct another equilibrium, this one with "centralized communi­
cation," analogous to the "Law Merchant System Strategy" of Milgram
et a!. Again, this equilibrium may exist under conditions when the no­
communication and multilateral-communication equilibria do not. It rep­
resents a formal institution: not only are certain acts of consultation and
reporting required of the players, but one player is designated to play a
special "official" role.
Finally, section 5 addresses the problem of providing positive incen­
tives for the "official" to eschew corruption, which is rendered possible
by the addition of sufficient opportunities for communication and pay­
ment. This demonstrates one aspect of how an institution must solve the
classical problem of "who watches the watchers"; additional provisions
for monitoring and sanctions would likewise have to be incorporated as
equilibrium behavior within the original game. Section 6 concludes with
remarks on the design or emergence of institutions, and on the applica­
tion of the equilibrium approach for the study of institutions and of
behavior within institutions.
2. Cooperation without an Institution
Consider first the simple setting in which a set of players {1,2, .. . ,N}
(N � 2 and even) engages in indefinitely repeated, pairwise plays of a
two-player prisoner's dilemma (PD) game, G, with one another. The
game G has the payoffs shown in matrix 1, where f3 > 0, a> 1, and a­
f3 < 2.8 Specifically, in each period t = 1, 2, . . . , players are paired by a
random mechanism in which the probability of any player i being paired
with a particular other player j is 1/(N- 1). Players i and j then play one
iteration of G. Players discount payoffs in future periods by some dis­
count factor o < 1. Each player has complete, perfect information about
his or her own past and present interactions, but no information about
interactions between pairs of other players. Let G(B) denote this entire
repeated game with random matching of players.
One equilibrium is of course for players to always defect uncondi­
tionally. There are other equilibria in which cooperation occurs. For
example, suppose that each player uses a subgame-perfect version of the
62 Explaining Social Institutions Rational Actors, Equilibrium, and Social Institutions 63

Player 2
The proof of the lemma, and of all subsequent theorems, is relegated to
the appendix. Note that y is a strictly increasing function of o. Table 1
c D indicates some representative values of y given 8 and N. Obviously,
Player 1
c 1' 1 -13. a increasing group size rapidly washes out the "shadow of the future"
( Axelrod 1981) necessary for cooperation.
D a, -13 0, 0

THEOREM 1. The strategy profile in which each player plays TFT with
Matrix 1. The 2-player PO game, G. each partner individually is a perfect Bayesian equilibrium of G(o) pro­
vided that 8 is sufficiently large. In particular, 8 must be greater than both

(N- 1)(a- 1) (N-

and 1){3
tit-for-tat strategy against each other player, regarding each series of (N- 2)(a- 1) + f3 + 1 (N- 1)/3 + 1 ·

pairwise interactions with the same partner as a separate game from the
interactions with other partners. That is, to the usual tit-for-tat strategy
we add the prescription that if for any reason a player does depart from
the strategy and defect, on the next two iterations ( here, the next two
encounters with the same partner) that player cooperates uncondition­
ally, making "restitution," so to speak, and thereafter resumes condi­
tional cooperative play.9 I shall refer to this version of the strategy as
"TFf." If the discount factor for each pairwise interaction, say y, is
sufficiently large, this would be an equilibrium. In that case, we have
what would normally be called a "norm of reciprocity" ( Axelrod 1981;
Calvert 1989): each player expects his or her opponent to reciprocate
cooperation and retaliate against defection in such a way as to make
defection not worthwhile. Such reciprocity is a norm in a broad sense,
although deviations are of no consequence to the group as a whole, but
only to the victim of that defection. By the definition suggested in sec­
tion 1, this equilibrium does not qualify as an institution.
A necessary and sufficient condition for cooperation to be possible
in this setting is thata s 11( 1- y). In that case, the strongest possible
punishment, the "grim trigger" strategy of permanent retaliation against
a partner who has ever defected, will deter defection even if it is only
carried out by the partner against whom the defection occurred. Alterna­
tively, TFf can be implemented provided that y exceeds a somewhat
larger bound; the following results formalize this fact in terms of a
condition on the original discount factor, o.
Table 2 shows the lower bound for 8 for several values of a, {3,
andN.
3. Communication and Cooperation:
An Informal Institution
As is apparent from table 2, if N is moderately large a healthy 8 may
be insufficient to support this cooperative institution in equilibrium.
Due to the rarity of repeated meetings between any two players and to
the lack of information about one's partner's behavior with other part­
ners, no strategy of reciprocation can be viable under these conditions.
Even in large groups, however, it may still be possible to support
cooperation in equilibrium· if there is more widespread punishment of
defectors. This could be done if the players could inform one another
about outcomes.
Accordingly, I revise the above model by explicitly allowing for the
possibility of communication among the players. Let G* be the stage
game played at each period t; G* consists of the following substages or
phases:
TABLE 1. Representative Values of y, the Discount Factor for Interactions with a
Particular Player, as a Function of the Overall Discount Factor Band the Number
of Players, N

LEMMA. In G(o), the proper discount factor for interactions with a 8 = .999 8= .99 8 = .95 8 = .90 8 = .75 8 = .50
particular other opponent is given by
N<=5 .996 .96 .83 .69 .43 .20
N 10
= .991 .92 .68 .50 .25 .10
0 N=30 .972 .77 .40 .24 .09 .03
y = N- 1 - o (N - 2) . N= 100 .910 .50 .16 .08 .03 .01
64 Explaining Social Institutions Rational Actors, Equilibrium, and Social Institutions 65

1. Players are paired at random ; each player knows only the pairing
that she herself is in.
2. Players may communicate with one another. Any player i may
send a message to any other player j, and messages may be sent
to as many other players as desired. For each player j whom she
contacts in this manner, player i bears a communication cost c.
All of these communications take place simultaneously. The mes­
sages must be chosen from some message set or "language" L.
3. Any player who received a message in phase 2 may deliver an
immediate, costless response also drawn from L.
4. The paired players play one iteration of G; again, one knows
only the outcome in one's own pair.
5. The players may again communicate as in phase 2, simulta­
neously sending messages chosen from L to as many other play­
ers as desired, bearing a cost c for each contact. A player cannot
verify to her opponent in the current phase what messages were
sent to other players.
Let G*(8) denote the repeated game thus defined. As will become appar­
ent, this temporal ordering of communication and play is chosen in
order to enable study of the particular equilibria here and in section 4 (in
fact, phases 2 and 3 won't be used until section 4).Notice that a strategy
consisting of the pairwise tit-for-tat of theorem 1, plus a prescription that
the player never communicate and ignore all communication from oth­
ers, is a subgame perfect equilibrium in G*(8). Likewise the presence of
additional communication phases will not affect theorems 2 and 3 below,
since any equilibrium in the game as specified here can be extended to
specify no sending of messages in the added phases, and ignoring of any
messages that are sent in violation of this prescription. Also, the "lan­
guage" could contain a large and arbitrary set of possible messages
without affecting whether a particular strategy profile is in equilibrium;
the equilibrium could be extended to specify that any message not from
the restricted language used for the equilibrium is ignored.
TABLE 2. Examples of Bound for Theorem 1
For present purposes, assume that the language L includes all pos­
sible pairs (n,a) where n E N and a E {C,D}; for this equilibrium, the
message (n,a) may be taken to mean, "My partner in this stage was n;
she played a." Now define the strategy of "tit-for-tat with multilateral
communication," TFT/MC, in which a player behaves as follows in the
phases of each stage:
2. Send no messages before play.
3. Make no replies.
4. If in cooperation status (defined below), play C if partner is
reported to be in cooperation status and D if partner is reported
to be in punishment status (also defined below); if in punish­
ment status, play C.
5. If in cooperation status, truthfully report partner's identity n
and action a to each of the N - 2 other players, incurring cost
(N-2)c; if in punishment status, do not communicate.
Any communication in phases 2 or 3 and any communication other than
the prescribed reports in phase 5 are ignored. A player begins G*( 8) in
"cooperation status." A player enters (or remains in) "punishment sta­
tus" if she fails to play as prescribed in phase 4, or if she fails to report
when prescribed in phase 5. A player in punishment status returns to
cooperation status immediately upon playing C in ph�se 4. Note that a
player may deviate from this strategy by falsely reporting her partner's
status in phase 5; if that happens, then a player in cooperation status
may be treated by other players as though she were in punishment
status. As the note on the proof of theorem 2 (see appendix) indicates,
such lying does not occur in equilibrium.
Just as in theorem 1, it is possible to specify a condition on 8 so that
this strategy is in equilibrium.
THEOREM 2. The strategy profile in which all players use the strategy
TFT/MC is a perfect Bayesian equilibrium of G*(8) for sufficiently large
8 provided that c is less than each of the following values:

alpha 1.1 1.1 1.1 2 2 2 2 4 4 1_ 2-a +{3 8

N - 1 (1 + {3).
_

beta 0.1 0.5 3 1.1 1.5 3 5 2.1

' and
3 N - 2' N - 2
N 2 0.091 0.333 0.750 0.524 0.600 0.750 0.833 0.968 0.750
5 0.286 0.667 0.923 0.815 0.857 0.923 0.952 0.992 0.923
10 0.474 0.818 0.964 0.908 0.931 0.964 0.978 0.996 0.964
The exact lower bound for 8 to support this equilibrium is the max of
30 0.744 0.935 0.989 0.970 0.978 0.989 0.993 0.999 0.989
100 0.908 0.980 0.997 0.991 0.993 0.997 0.998 1.000 0.997 (a- 1) + (N - 2)c and {3 + (N - 2)c
{3+1 {3+ 1
66 Explaining Social Institutions Rational Actors, Equilibrium, and Social Institutions 67

This combination of multilateral communication, cooperation, and retal­ described as follows.When paired with player 1, always defect and never
iation is properly an institution because failure to cooperate and failure send any messages (the idea being that, for simplicity, the director re­
to report are punished by every partner encountered until "restitution" frains from actual play of the PD game). When paired with any other
is made.That is, any player's actions today affect every player's actions player j in {2, . . . , N}, player i observes the following prescription for
tomorrow.This institution has minimal organizational complexity, how­ the respective phases of each stage in the game:
ever, since every player has exactly the same role in communicating and
in punishing deviation. Nevertheless, notice bow the "rules of the game" 2. In stage t = 1, do nothing. In stages t > 1, if in punishment
for these players are much more complex than in the pairwise TFf status (defined below), do nothing; if in cooperation status (de-
equilibrium of theorem 1: beyond simply interacting with each partner, a fined below), then pay c to send message Qi to the director,
player is expected by the whole group to act in a certain way and to where j is i's current partner.
communicate certain information at certain times to certain other peo­ 3. Do nothing.
ple. Any failure to act or communicate as expected incurs punishment. 4. In stage t = 1, play C. In stages t> 1, when in cooperation status
However, these rules are not part of the structure of the underlying and told (by the director's reply in phase 3) that i is in coopera-
game; rather, they are features of the equilibrium strategy profile. If tion status, play C. Otherwise play D. When in punishment
discounting is too heavy or communication cost too high, such a system status, play C (make "restitution").
of rules is impossible to maintain.
Notice moreover that the bound on o is generally much lower using TABLE 3. Examples of Bound for Theorem 2
TFT/MC than it was in the more anomie system of TFT. Table 3 shows
the bound for the same values of a, {3, and N as in table 2, for various Small c (c = .001)

values of c: a minimal .001 in the first part, pushing c's upper bound for alpha 1.1 1.1 1.1 2 2 2 2 4 4
N = 100 in the second part, and even higher, pushing the bound for N = beta 0.1 0.5 3 1.1 1.5 3 5 2.1 3
30 in the third part. Especially for low values of a or high values of {3, N 5 0.094 0.335 0.751 0.525 0.601 0.751 0.834 0.969 0.751

even fairly costly communication makes it much easier to achieve co­ 10 0.098 0.339 0.752 0.528 0.603 0.752 0.835 0.970 0.752
30 0.116 0.352 0.757 0.537 0.611 0.757 0.838 0.977 0.757
operation than it was using TFT.
100 0.180 0.399 0.775 0.570 0.639 0.775 0.850 0.999 0.775

4. Cooperation Using a Formal Institution c at .0001 below bound for N = 100

alpha 1.1 1.1 1.1 2 2 2 2 4 4

Since communication is costly, its use in the equilibrium just described
beta 0.1 0.5 3 1.1 1.5 3 5 2.1 3

N
reduces the resulting gains from cooperation. Each individual must bear a c 0.010 0.010 0.010 0.010 0.010 0.010 0.010 0.001 0.010
communication cost of ( N - 2)c on each iteration, so the communication 5 0.119 0.354 0.758 0.538 0.612 0.758 0.838 0.969 0.758
cost in a large or far-flung group may outweigh even large cooperation 10 0.165 0.388 0.770 0.563 0.633 0.770 0.847 0.970 0.770
gains.In such a group, cooperation would require a more efficient ex­ 30 0.351 0.524 0.821 0.660 0.714 0.821 0.881 0.977 0.821
100 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000
change of information.As we will see, such an efficiency gain can be had,
but only by achieving a greater degree of organizational complexity. c at .0001 below bound for N = 30
In place of the multilateral communication used in TFf/MC, I con­
sider next a scheme of centralized communication.10 Arbitrarily desig­ alpha 1.1 1.1 1.1 2 2 2 2 4 4
beta 0.1 0.5 3 1.1 1.5 3 5 2.1 3
nate player 1 as the "director," who will serve as a central clearinghouse
c 0.036 0.036 0.036 0.036 0.036 0.036 0.036 0.004 0.036
of information. Assume now that the language L includes at least the
N 5 0.188 0.405 0.777 0.575 0.643 0.777 0.851 0.971 0.777
elements (n,a) as before, plus the messages Qi for each j EN, which for 10 0.351 0.524 0.821 0.660 0.714 0.821 0.881 0.977 0.821
purposes of this equilibrium may be interpreted as the query, "My oppo­ 30 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000
nent is j; what is his status?"11Each player i in {2, 3, .. . , N} follows 100 X X X X X X X X X
a strategy of "tit-for-tat with centralized communication" (TFf/CC),
Nol�: x den otes that c exceeds its upper bound for these parameter values.
68 Explaining Social Institutions Rational Actors, Equilibrium, and Social Institutions 69

5. If in cooperation status, pay c to report j's action (j,a), to the
director. If in punishment status (i.e., just played D inappropri­
ately in phase 4) then do nothing.
Player 1, the director, obeys strategy A, described as follows. For all t,
in phase 2, make no statement; in phase 4, always defect. In phase 5,
make no communication. Otherwise:
3. In stage t = 1, say nothing. In stage t > 1, if message Q1 was
received from player i in phase 2 , truthfully report the status of
that player's reported opponent - (j, C) if cooperation, (j,D) if
punishment- and otherwise communicate nothing;
The cooperation and punishment statuses are defined as follows. At
stage t = 1 , every player is in cooperation status. A player i in coopera­
tion status enters punishment status if any of the following occur: in
phase 2, she fails to query as required; in phase 4, she unilaterally fails
to cooperate with a partner reported by the director to be in coopera­
tion status; or if she fails to report as prescribed in phase 5. If a player in
punishment status cooperates in phase 4, she reenters cooperation sta­
tus beginning with phase 5.
The following theorem derives the conditions under which the
profile TFT/CC can be used by all players (except the director) in
equilibrium:
different players play different roles. As the second part of table 4
demonstrates, even if the communication cost is so high that it burns up
50 percent of the gains from cooperation, the institution with central­
ized communication compares favorably with TFI in its ability to sup­
port cooperation in large groups. Note that table 4 shows larger group
size to be an advantage. This happens because higher numbers mean
that a player is paired with the director (and thus gets a zero payoff) less
often. Of course, if larger group sizes put more strain on the director's
ability to cope with her duties, this advantage might disappear.
In the TFT/CC equilibrium, we have an institution that is actually rec­
ognizable as an organization. Despite the fact that the underlying game is
the same as those used above to illustrate TFT and TFT/MC, this institu­
tion defines a special role for one player, forcing the other players to report
only to the director and to base their actions on only the director's replies.
Moreover, the institution creates unique "actions" that the director can
take, namely the accepting of queries and the sending of messages concern­
ing the status of players. This apparent addition of new strategies to the
game, however, is akin to the apparent introduction of new rules discussed
after theorem 2: the underlying game really remains constant, and only
the equilibrium has new features. A player could have departed from
TFT/MC and reported other players' statuses all she wanted under the
equilibrium of theorem 2, assuming that those messages were in the lan­
guage; but such reports would not, in that equilibrium, have affected the
TABLE4. Examples of Bound for Theorem 3

THEOREM 3. The strategy profile in which player 1 uses A and players

2 through N use TFT/CC is a perfect Bayesian equilibrium of G*(o) for
c at .0 1 below Theorem 2's bound for N = 100

sufficiently large o provided that c < V2 and c < 1 - (a - {3)/2. The lower alpha 1.1 1.1 1.1 2 2 2 2 4 4
bound on o is the maximum of beta 0.1 0.5 3 1.1 1.5 3 5 2.1 3
c 0.010 0.010 0.010 0.010 0.010 0.010 0.010 0.001 0.010
N 5 0.130 0.410 0.804 0.602 0.673 0.804 0.872 0.976 0.804
{3 + c (N - 1)(cx- 1 +c) 10 0.112 0.369 0.776 0.560 0.634 0.776 0.852 0.972 0.776
N - 2 and
(N - 3)(/3 - c + 1) + a + {3
30 0.104 0.350 0.761 0.540 0.615 0.761 0.841 0.969 0.761
{3 + c + (1 - 2c) N 1
_
100 0.102 0.345 0.756 0.534 0.609 0.756 0.838 0.969 0.756

Table 4 gives representative values of this lower bound for the same Very large c (c = .25)
values of ex, {3, and N as in tables 2 and 3. Part 1 uses the same values of alpha 1.1 1.1 1.1 2 2 2 2 4 4
c as in table 3, Part 2; the second part uses a larger value for the
communication cost. Comparing Part 1 with table 3 shows that, in cases N
beta
5
10
0.1
0.483
0.441
0.5
0.667
0.628
3
0.897
0.880
1.1
0.783
0.752
1.5
0.824
0.797
3
0.897
0.880
5
0.933
0.922
2.1
X
X
3
0.897
0.880
where N and care large, the formal institution using centralized commu­
30 0.420 0.608 0.871 0.737 0.784 0.871 0.916 X 0.871
nication makes cooperation possible under conditions in which the infor­
100 0.414 0.602 0.868 0.732 0.780 0.868 0.914 X 0.868
mal norm using decentralized communication could not be maintained.
Centralized communication is much more efficient, but requires that note: "x" denotes thai c exceeds its upper limits for these parameter values.
70 Explaining Social Institutions
actions of other players. Likewise, players other than the director could
make such statements by departing from TFf/CC, but other players wi11
react only to the director. The TFf/CC institution presents the players
with rules that they must, out of their own self-interest, follow; but a
change in the game's parameters, such as a lowering of the discount
factor, could render those rules ineffective and unenforceable.
One could view the director as a third-party enforcer in this model,
since the director in effect pronounces a sentence on the deviant player,
a sentence that will then be carried out by rational players. I n a more
complex model, one could create even more explicit equilibrium punish­
ment schemes, in which a player is designated as "enforcer" and whose
instructions to punish are obeyed by a set of deputies. This kind of
internal enforcement differs critically from the "third-party" enforce­
ment often used to explain cooperation in collective-action models,
however. The designation of, actions of, and obedience to the enforcer
must be made in the players' interest for such a system to be in equilib­
rium. The simple model of TFf/CC in G*( o) illustrates exactly how
such a system can be specified. Likewise it should be possible to identify
equilibria in more complex models in which a designated "leader" gives
"orders" that are "obeyed" by the players, and whose disobedience is
punished. Again, all of the required actions and sanctions must be made
compatible with rational behavior by all players, including the director,
who must be willing to order the actions when appropriate, and willing
to forbear punishment except when appropriate, an issue to which I
turn in the next section. This, I argue, is how real-life institutions work.
The reader may have noticed that I have said nothing about how
such institutions may come to exist. This is a nontrivial question since
there are many possible institutions even for such a simple model as
G*( 8), particularly when communication is allowed, and even if the play­
ers all expected to use TFf/CC they would still have to agree on who was
to be the director. I return to this issue in the concluding section.
5. Closing the Model: Who Directs the Director?
The model so far omits two important possibilities: that the director
might not be willing to serve, or that the director might engage in corrup­
tion. To bring closure to the model of institutions as equilibria, it is
necessary to provide the necessary inducements and deterrence within
the game itself. Suppose that the director collects a fee ¢ each time any
player communicates with the director in phases 2 or 5. Then the total
cost to a player of communicating with the director becomes d = c + ¢
(where cis the inherent cost of communication, as in the model of sec-
Rational Actors, Equilibrium, and Social Institutions 71
tions 3 and 4), and each transaction between players in cooperation status
using the equilibrium strategy yields the director a payoff of 4¢. To make
players indifferent between serving as director and being active players,
set ¢ so that the total payoff from serving as director equals that from
being a player: in terms of per-period payoffs, 4¢(N - 2)/2 = (1 - 2d)
(N - 2)/(N 1), that is,
-
1 - 2c
¢
= -u::;-·
Naturally, for TFf/CC to be an equilibrium, all the conditions in theo­
rem 3 must hold for d and not just for c; but with a large group, the
difference ¢ will be small.
This leaves the issue of corruption. If the game were to incorporate
opportunities for additional communication and for transfers of wealth
among players, it would become possible for the director to extort
money from a player under the threat that, if the player did not pay the
extortion money, the director would falsely report him as being in pun­
ishment status. Or, a player in punishment status could bribe the direc­
tor to falsely report that he was in cooperation status. Several methods
may be used to prevent such corruption. First, the players could moni­
tor the director and apply some form of punishment to a director discov­
ered to be corrupt. This approach would impose monitoring costs, and
would require additional institutional structure (i.e., a more detailed
specification of equilibrium behavior in an appropriately modified ver­
sion of the game) to carry out the monitoring and punishment. Second,
the availability of several directors among whom players could choose
by mutual consent could suppress corruption through competition.
Third, the refusal of players to continue to cooperate when a director
committed extortion would threaten a corrupt director with the loss of
the fee from successful transactions.
In the remainder of this section, the latter approach, resembling
that used by Milgram, North, and Weingast to achieve a similar goal, is
used in a modified version of the model to demonstrate how to discour­
age corruption. More elaborate means would be necessary to minimize
all forms of corruption, and the costs of these means would add further
conditions on the possibility of cooperation. This simple model, how­
ever, will serve to demonstrate all of the important building blocks of
institutions among rational actors.
To make it possible for the director to extort the players, add a
second communication phase 3a, which immediately follows phase 3.
72 Explaining Social Institutions
We now assume that it is possible for the director, in response in phase 3
to a player's query, to extort the player rather than answer the question:
only in return for an immediate extra payment B in phase 3a (for which
there is no additional communication cost) will the director report the
player to be in cooperation status. If the payment is refused, the direc­
tor will falsely report the player to be in punishment status. I assume
that this threat is credible, and demonstrate how it can nevertheless be
deterred in an equilibrium that in all other respects is subgame perfect.
In the new phase, 3a, the extorted player may either pay the director
the amount B demanded, or else pay a communication cost c to warn his
partner that the director has acted corruptly. Another extra stage, 3b, is
added so that the director may reply to the queries, truthfully or other­
wise. Players are assumed to be unable to distinguish, simply from the
timing of replies, between corrupt and honest behavior by the director.
To identify an equilibrium that supports cooperation while prevent­
ing extortion, suppose that the director now follows strategy A*:
3. If a querying player has ever before acquiesced to an extortion
attempt and is paired with a partner other than one cheated in
the immediately preceding iteration, demand a payment a - d
- ¢/(N - 3), unless a larger payment has ever been demanded
of a player who had acquiesced at some earlier time; in the
latter case, demand a payment of a. Otherwise, take no action;
3b. If a player has refused an extortion attempt, and his partner has
queried, report the player as being in punishment status; other­
wise, if a player defected or failed to report on the previous
iteration due to an attempt to extort one or both players, then
report that player to be in cooperation status; otherwise, report
honestly as in strategy A;
and otherwise follow A. Suppose further that players 2 through N fol­
low the "antiextortion strategy" (AES). This strategy modifies TFT/CC
by stipulating that any player who fails to pay the director's cf> enters
punishment status, and by adding the following instructions:
2. A player who has entered permanent punishment status (see in­
structions for phase 5 below) does not report; otherwise, report;
3a. A player who is extorted and who has made any extortion
payment in the past pays any amount demanded up to and
including a, unless in permanent punishment status (but if a
higher payment is demanded, or if in permanent punishment
status, pays nothing); a player who is extorted, and has never
Rational Actors, Equilibrium, and Social Institutions 73
acquiesced to extortion previously, refuses any payment and
communicates to her partner (at cost c) that extortion has been
attempted;
4. If an extortion attempt has been reported by either player, then
D; if no extortion attempt was made and none has ever been
paid, follow TFT/CC. If an extortion payment was demanded,
or no payment was demanded but extortion has been paid
previously, then D;
5. If an extortion attempt was reported by either player in (3a)
and both players played D as a result in phase 4, then do not
report; if no extortion attempt was made and none has ever been
paid, follow TFT/CC. A player who has ever made an extortion
payment does not report. If an extortion demand of more than
a - d -cf>I(N - 3) was made in (3a), enter permanent punish­
ment status (never report or cooperate in any future iteration).
Under this strategy, no extortion should ever take place, and if it does
no payment should ever be made. The final result, then, is:
ThEOREM 4. The strategy profile in which player 1 uses A * and players
2 through N use AES is a perfect Bayesian equilibrium (even assuming that
the director can commit to his extortion threat) for sufficiently large o,
provided that the other conditions oftheorem 3 are met with d in place ofc.
Obviously the conditions of theorem 4 are slightly more stringent than
those of theorem 3 : besides supporting the increased communication
cost d, it must also now be rational for the extorted player to resist.12
This is the kind of closure required to really explain an institution.
Notice that the equilibrium defines an organizational form, puts an
individual in a position of authority, mandates and carries out enforce­
ment (in this case decentralized), and thus "constrains the behavior" of
the individuals involved. The institution thus defined requires no out­
side enforcement, nor does it depend on any irrational commitment to
observe or enforce the equilibrium. Rather, the institution is no more
than a regular behavior pattern sustained by mutual expectations about
the actions that others will take when anyone violates the rules ­
whether the violation be defection, lying, nonpayment of fees, extor­
tion, or malfeasance in office. The institution is just an equilibrium.
6. Conclusions
Using a modified version of the repeated two-player PD game, I have
demonstrated various kinds of equilibria and the conditions under
74 Explaining Social Institutions
which they can support cooperation. These range from a collection of
unconnected two-person reciprocity arrangements that, taken together,
would not qualify as an institution, to a complex combination of expecta­
tions, required communication, and formal roles that serve to provide
monitoring and sanctions and to ensure against malfeasance. For large
groups or groups with significant costs of communication, the formal
institution makes cooperation possible while the series of informal un­
derstandings does not.
Another way to restate the point of this whole exercise is that there
is, strictly speaking, no separate animal that we can identify as an institu­
tion. There is only rational behavior, conditioned on expectations about
the behavior and reactions of others. When these expectations about
others' behavior take on a particularly clear and concrete form across
individuals, when they apply to situations that recur over a long period
of time, and especially when they involve highly variegated and specific
expectations about the different roles of different actors in determining
what actions others should take, we often collect these expectations and
strategies under the heading of institution. This is not to say that institu­
tions do not exist. Rather, it is to say that there are no institutional
"constraints" or "preferences" aside from those arising out of the mu­
tual expectations of individuals and their intentions to react in specific
ways to the actions of others, all in an attempt to maximize utility in a
setting of interdependency. Institution is just a name we give to certain
parts of certain kinds of equilibria.
The idea of institutions as equilibria can be fruitfully applied to
analyze the stability and effects of institutions in various areas of social
life. Such an analysis would begin by portraying the game that underlies
a given institution: the production of value, the opportunity for gains
from trade, cooperation, coordination, or whatever. Along with this
basic interaction, most institutions depend upon information transfer,
and so opportunities for communication must be built into the game.
The institutions and behavior observed in real life should constitute an
equilibrium to the game so constructed. An examination of how the
model's equilibrium behavior varies with variations in the game's pa­
rameters (payoffs, number of players, discount factors, etc.) then pro­
vides predictions about how marginal changes in the real-world environ­
ment will influence real-world behavior. It may be possible to identify
parameter changes under which the observed institutional features will
cease to be in equilibrium (as happens in the hypothetical institutions
modeled in the previous section if the discount factor drops below a
certain value). Further, interrogation of the model yields results con­
cerning Pareto improvements that might be available under alternative
Rational Actors, Equilibrium, and Social Institutions 75
equilibria, or strategic opportunities for coalitions. Empirical analysis of
such a model would consist of tests of those comparative statics results,
along with the detection of infrequent behavioral patterns predicted by
the model but not previously noticed in the real-world case at hand.
Looking at institutions as equilibria, it is possible to understand
institutions themselves in terms similar to those in which we understand
rational action within them. One can describe institutions that are stable
at one time but change or fall apart at another, without appealing in one's
analysis to exogenous forces other than the revelation of information in
the course of play or some exogenous parameter shift. Such an analysis
generates predictions for the stability and instability of the institution in
the form of theoretical responses of player behavior to the parameters of
the game. In short, this approach is a useful tool to guide empirical study
of institutions. It can clarify the nature and workings of real-life institu­
tions either through formal game-theoretic modeling or through informal
application of the principles.
In the two sections that follow I illustrate this approach, first with a
new informal application to an existing controversy, and then with inter­
pretations of several analyses already in the literature. The section follow­
ing that discusses the problems of institutional formation, and of their
functionality or dysfunctionality for the welfare of their participants.
Congressional Committees and the ex Post Veto
The exchange between Shepsle and Weingast (1987) and Krehbiel
(Krehbiel, Shepsle, and Weingast 1987) concerning the foundation of
committee power and the control of interchamber conferences to resolve
bills in the U.S. Congress provides a nice illustration of how an informal
application of the institutions-as-equilibria approach can clarify our think­
ing about a specific political institution. Shepsle and Weingast, using a
pure institutions-as-constraints analysis, conclude that committees' con­
trol of the introduction of bills together with their ability to kill a bill at
the conference stage if it is amended beyond their liking, the ex post veto
power, enables them to extensively control legislation within their respec­
tive jurisdictional areas. In particular, Shepsle and Weingast discount
norm-based theories of "deference to committees" to account for appar­
ent committee power, arguing that there are insufficient means to protect
such informal reciprocal agreements from the ubiquitous opportunities
for "reneging and opportunistic behavior. "13 Instead, committee power
must be accounted for by more formal means of enforcement, and the ex
post veto is key among them. According to Shepsle and Weingast, this ex
post veto power derives from committees' nearly automatic ability to
76 Explaining Social Institutions
supply members for conference committees, and from standard floor
procedures that effectively constrain the House and Senate, when voting
on conference committee reports, to a simple up-or-down vote.
Krehbiel criticizes this view, arguing that committees hold no such
power because a majority in either chamber (he concentrates on the
House, as do Shepsle and Weingast) can easily override any committee
attempt to exercise a conference-stage veto by using other procedures at
their disposal, such as instructing their delegates to the conference com­
mittee, bargaining directly with the other chamber without going to
conference, or discharging the conferees and reclaiming the bill.
Shepsle and Weingast respond, in part, by pointing out that on major
bills the process nearly always operates as they have described, and that
there are occasional instances (at least during the strong committee era
preceding the 1970s) of conference committees killing bills in a manner
consistent with the ex post veto story. This indicates, say Shepsle and
Weingast, that committees effectively have the ex post veto power even
though methods of circumvention are available. Presumably, they con­
clude, the cost of circumventing the conference committee on a major
issue is usually prohibitive. However, they leave vague the nature of
these costs, which are unobservable and, as they and Krehbiel all point
out, can only be inferred after the fact, when we can observe the use or
disuse of the methods of circumvention.
The institutions-as-equilibrium approach provides a clearer view of
the nature of such process-based power, giving a solid theoretical basis
to the "cost" that holds conference procedures in place, and forcing a
modification of Shepsle and Weingast's view of the inefficacy of infor­
mally enforced agreements.I4 Consider the problem that conference
procedures are created to address. When the two chambers pass bills
that differ in many respects, and that elicit a whole spectrum of opinion
on each point of disagreement in each chamber, the problem of negotiat­
ing a compromise bill is a messy one. If not managed correctly, the
negotiation promises to be time- and effort-consuming and has a large
probability of failing; meanwhile, other goals of the legislators involved
in the negotiation go unaddressed. In any such bargaining process,
efficiency requires that an acceptable offer be generated as quickly as
possible; even though that offer may be better for some members of a
majority than for others, its quick acceptance would then clear the way
for all of them to consider other matters and address other goals, includ­
ing those of members whose needs were not completely met by this bill.
In short, this is a garden-variety problem of coordination with "mixed
motives" (Schelling 1960).
Congress has developed several methods for conducting this bar­
gaining that correspond to Krehbiel's list of circumvention methods
Rational Actors, Equilibrium, and Social Institutions 77
(Krehbiel, Shepsle, and Weingast 1987, 930-31). One chamber can
simply accept the bill passed by the other chamber, a useful device when
there is little disagreement between some pair of majorities. When
there is moderate disagreement, the chambers can "message" a bill
back and forth, each either insisting on its own provisions or concurring
in the other's. This is a more ponderous method, since each acceptance
or counteroffer must be agreed on through a parliamentary process
involving the whole chamber. The conference process offers a more
streamlined way to generate a compromise that majorities in both cham­
bers can live with: negotiation is conducted by a small group of confer­
ees, and the proposal typically gets only an up-or-down vote in the two
chambers. If either (1) the conferees represent a wide range of opinions
from each chamber, or (2) members of each chamber are willing (in the
expectation of similar future considerations) to allow their conferees
wide latitude, then the conference process can generate an acceptable
proposal much more easily and quickly than could messaging between
the two chambers.
In order for a chamber majority to "roll," or amend the report of, a
conference committee, it is necessary to put together the necessary
voting coalition in one's own chamber to agree on an alternative also
acceptable to some majority within the other chamber. This is another
complicated bargaining problem, and if the original conference report is
not too bad, attempting to amend it may just not be worth the trouble;
aside from the costly effort involved, there is again usually a high proba­
bility of failure. As a result, the members of the conference committee,
since they have some leeway to propose various compromise bills, can
propose ones to their own liking. If the original conference committee
fails to reach an agreement (even if "on purpose," as the Shepsle­
Weingast ex post veto threatens), it may likewise not be worth the
trouble to put together an alternative conference committee or to use
more cumbersome means to produce a compromise bill.
Notice that the appointment of conferees is itself a bargaining or
coordination problem in the chamber, given that the conference proce­
dure itself is in place. In the absence of some standardized, generally
accepted method (such as the Speaker appointing members of the com­
mittee and subcommittee of original jurisdiction), identifying a group of
conferees acceptable to a majority may again cost more than it is worth
in terms of time and effort forgone for other purposes. The result is that
the people who get to select the point of coordination (here, the confer­
ees) have a measure of real power relative to other participants. is
In short, if for present purposes we take other legislative rules and
procedures as fixed, the "institution" of the committee-dominated con­
ference process is an equilibrium (and a welfare-enhancing one, at least
78 Explaining Social Institutions
if you are often in the majority) in the underlying coordination game of
reaching agreement between the chambers. This view of the process
identifies the cost of overriding the conference committee as a function of
the difficulty of the coordination problem that members must face if they
pursue an override. More extensive disagreement between and within
chambers makes an override more costly. For example, the presence of
organizational alternatives to the conference committee, such as a preex­
isting, informal "caucus" with many members and similar goals on the
bill, makes it easier and therefore Jess costly to roll the conference com­
mittee. When the value of opportunities forgone due to the override
attempt is high, such as late in a Congress in which many important bills
remain to be considered, the override is more costly. Overall, the trick
for conference committee members is to moderate the extent to which
they change (or eliminate) the content of bills so as to stay within the
limits that define whether rolling the committee is worthwhile. True to
Krehbiel's argument, there are clearly limits on the ex post veto power.
True to that of Shepsle and Weingast, however, standing committees
would be weaker if they did not have this opportunity to modify the bill
after it leaves the floor, and did not benefit from the chamber majority's
inclination to accept the conference report as a coordination solution.
A more formal model would have to include a more elaborate specifi­
cation of the game underlying the legislative process, and would address
a wide array of procedural rules than just the conference provisions, since
ramifications of flouting the conference committee could spill over into
other areas oflegislative politics. The maintenance of an effective confer­
ence procedure is a collective good for most members of Congress; mem­
bers may therefore design a system in which respect for the conference
process can be enforced partly by sanctions outside that process, such as
by a committee in the consideration of a later bill, or by party leaders in
the awarding of valued perquisites (in which case a stronger legislative
party organization ought to cause stronger conference committees). The
institutions-as-equilibrium approach views all such legislative rules,
norms, and prerogatives as being enforced within the context of equilib­
rium in the underlying game. Formal rules and unwritten norms alike
must ultimately be enforced through the expectations and rational behav­
ior of members. This view calls into question, then, the broad argument
of Shepsle and Weingast that informal agreements are unsustainable due
to the nature of legislative interactions. I would argue instead that all rule
enforcement is endogenous, and that, just as the maintenance of infor­
mal agreements is always problematic, so is the enforcement of formal
rules. Perhaps Shepsle and Weingast's claim should be modified to say
that it is harder to enforce particular agreements about legislative specif-
Rational Actors, Equilibrium, and Social Institutions 79
ics than to enforce broad agreements about legislative procedure. Such a
claim, however, would require much more empirical or theoretical sup­
port than it has accumulated in the literature thus far.
Finally, the institutions-as-equilibrium model increases the ana­
lyst's ability to conduct "comparative statics" analysis of institutions, a
function that Shepsle and Weingast identify as a key useful feature of
their analysis. Their suggestion is that new procedures introduced begin­
ning in the early 1970s weakened the ex post veto consistent with
Krehbiel's general thrust, and that the effect on legislative outcomes of
such rule changes should be predictable using the ex post veto model as
a starting point.16 The institutions-as-equilibria model can go much fur­
ther, however, by providing a direct model of the underpinnings of the
institution to subject to comparative statics analysis. Such analysis could
treat institutional change in a much less ad hoc fashion than is possible
with the Shepsle-Weingast model. In particular, the present model of­
fers predictions on the relationship between the strength of the ex post
veto and various prior, observable correlates of override costs.
Some Existing Studies of Real Institutions as Equilibria
Actually, a few analyses of this general type have already appeared in the
literature, although they have probably not been recognized by most
readers as a genre. The brief examples used by Schetter (1981, chaps. 1-
2) to introduce his "economic" theory of institutions are the first of which
I am aware. The first extended analysis is that of Milgram, North, and
Weingast (1990), who study the Law Merchant, an institution for long­
distance trading in eleventh-century Europe. Milgram et al. construct a
stylized model of trading in which numerous merchants deal exactly once
with each partner, so that deterrence of cheating requires some method
for enforcing sanctions across interactions. The TFT/CC and AES institu ­
tions above are close relatives of the Law Merchant system that Milgram
et al. derive. The analyses by Greif, Milgram, and Weingast (1990) of
merchant guild organizations, and by Greif (1991) of Genoese and
Maghribi traders' use of trade agents, take similar approaches.
In her study of the changing institutional structure of the U.S. Sen­
ate, Sinclair (1989) analyzes norms by treating them as cooperative equi­
libria in a repeated PD, and informally examines the properties of those
equilibria as external conditions change, exactly as the institutions-as­
equilibria approach suggests. In a series of more formal analyses of
the impact of parliamentary procedure in the legislature, Baron and
Ferejohn (1989) and Baron (1989, 1991) stylize the process of recogni­
tion, proposal, and voting on the floor as a specialized type of bargaining
80 Explaining Social Institutions
game. Important features of formal and informal legislative practices
such as conditions governing the breadth of distribution of benefits
among the members; conditions under which amendments will or will
not be proposed under an open rule; and conditions under which mem­
bers prefer closed or open rules appear as equilibrium behavior in this
game. In other words, there is an underlying game of recognition, pro­
posal, and majority-rule voting (whose rules are assumed) that members
of a legislature must play; within that, they may adopt various recogni­
tion rules, amendment rules, and strategies, governed by equilibria that
depend on the parameters of the game. Similarly to my PD illustration
and to Sinclair's (1989) analysis of Senatorial norms, then, certain institu­
tional features may persist as equilibrium behavior patterns in the under­
lying game, depending on conditionsP
The Emergence and I n itiation of Institutions
The institutions-as-equilibria approach as presented in this chapter says
nothing directly about the emergence of institutions. Indeed, the tradi­
tional equilibrium approach explicitly ignores such questions, assuming
in effect that the players arrive unanimously at identical equilibrium
expectations through a priori reasoning. However, this leaves several
important matters to be resolved: how likely is a given institution (equi­
librium) to arise in real life? Can it rise spontaneously or does it require
explicit specification and agreement by members of the group? If it is
formed by explicit agreement, what considerations govern the group's
choice among possible institutions?
The spontaneous emergence of an institution is fundamentally a
problem of coordination. In the model examined in this paper, several
different institutions are derived, and many more are possible; even
within the TFT/CC definition, there is the question of which player
comes to be agreed upon as the "director." The problem for the players is
to arrive together at one such equilibrium. Still, the institutions in this
chapter are not all that complex, and over an extended period of play it is
reasonable to expect that similar institutions could arise by trial and error
with a minimum of purposeful planning. A good example of the partly
spontaneous emergence of an institution of similar complexity might be
the gradual formation of stock exchanges such as the London Exchange
in the early eighteenth century (Hirst 1948) and the American Exchange in
the late nineteenth (Sobel 1970) as central locations for the facilitation of
trade in bonds and corporate shares. Only after the initial, decentralized
emergence of securities trading in a central location did the rules of the
Exchange become matters of authoritative, centralized decision making.
Rational Actors, Equilibrium, and Social Institutions 81
Recently, new game-theoretic techniques have been suggested to
explicitly address the emergence of equilibria in games. In the "evolu­
tionary" or "learning" approach, for example, the mix of strategies used
by the set of players evolves through a process of imitative learning
or natural selection to a stable equilibrium state (see, for example,
Maynard Smith 1982 and Selten 1983); the set of such equilibrium states
is closely related to the set of Nash equilibria, and the well-behaved mem­
bers of either set generally tend to belong to both.18 Another approach,
due to Crawford and Haller (1990), models the process of learning to
coordinate through trial and error in a game of pure coordination. Such
models, building on a foundation of game-theoretic equilibrium analysis,
hold out the hope of more explicit models of the spontaneous emergence
of institutions. In all of them, Nash equilibria are, generally speaking, the
ultimate outcomes; but the choice among equilibria depends on initial
conditions, on the assumptions about the dynamic process, and on the
realizations of random variables such as when mixed strategies are played
out or when moves by "nature" take place.
Hechter (1990a) contrasts the latter approaches, which he labels
"invisible-hand" theories of institutional formation, with more purpose­
ful design of institutions through "social contract." Should players explic­
itly agree on a particular equilibrium of the underlying game as an institu­
tion, and then in some sense end their communication about institutional
design, they will have the proper incentives to adhere to the agreement
since it is an equilibrium. We might thus portray the process of creating an
institution as a bargaining problem, in which participants anticipate the
effects on them as individuals of each possible (equilibrium) institutional
arrangement. Any agreement reached is then automatically enforced
(since it is self-enforcing), as required for a bargaining problem. All of
the potential problems of bargaining, such as incomplete information
about other players' preferences, could then intervene to forestall the
reaching of any agreement. Alternatively and equivalently, a prominent
and acceptable suggestion of an equilibrium institution by a leader, in the
absence of coherent multilateral communication among players, can pro­
vide a "focal point" (Schelling 1960) that all players find advantageous to
follow. Obviously combinations of these two processes are possible as
well- consider the U.S. constitutional convention of 1787, in which a
small group of prominent men thrashed out an agreement that a wider
group of citizens then agreed to support.
Hechter argues that interesting real-world institutions more often
emerge through contract than via the invisible hand. While this may be
true (I remain agnostic, and in any event it depends on the universe of
institutions being examined), in real-world negotiations it is generally
82 Explaining Social Institutions
impossible to specify all contingencies in such a contract. Much is either
left up to customary interpretations (that often have emerged spontane­
ously) or determined along the way through a trial-and-error working
out of the contract's unspecified details. The development of the U.S.
Constitution is a perfect example of such a process. It is the quintessen­
tial social contract, but many hugely important details of that institution
have been worked out on the fly through presidential interpreta­
tions, court decisions, and public opinion. Thus, in contemplating the
development of institutions as equilibria, it pays to consider the nature
of spontaneous emergence.
The cooperative equilibria derived in this chapter could be imag­
ined either to have evolved through trial and error or to have resulted
from some initial communication phase in which the players made sug­
gestions about how to proceed. However, the literature still lacks an
explicit model of the evolution of equilibria in repeated games, where
the space of possible strategies is much more complicated than in a
simple normal-form game. A demonstration of the evolution of equilib­
ria in that context would make it more plausible that institutions as
equilibria could actually emerge in the real world.
It is important to keep in mind that institutions, whether designed or
spontaneous, can be dysfunctional as well as functional. A small coalition
might agree to cooperate in such a way that they can suppress by violence
the efforts of other players to obtain higher payoffs, creating an institu­
tion that makes the vast majority worse off than they would be without
any organization at all. Or, an evolutionary process can result in a Pareto­
inferior equilibrium, such as in the example of parents choosing the sex or
height of their children (Schelling 1 978, chap. 6), where the pressure to
be just slightly different leads to an extreme outcome making all people
worse off than if they had sought no advantage. Presumably, when groups
design institutions they attempt to make at least some members of the
group better off; and evolutionary processes among groups of strategic
and communicative players may be relatively easy to steer away from
dysfunctional results. However, the theory of institutions as equilibria
says nothing in general about the functionality of institutions, consistent
with the variety of welfare effects seen to flow from real-life institutions.
Conclusion
Schotter's (1981) "economic" treatment of social institutions underlies
the general idea presented here, that institutions are equilibria. This
chapter has developed and specialized Schotter's idea, applying it to
"underlying games" involving problems of cooperation and opportuni-
Rational Actors, Equilibrium, and Social Institutions 83
ties for communication. The generality and parsimony of this approach,
along with its ready applicability to specialized empirical analysis as dem­
onstrated by Milgram et al., Baron, and others, should make it a primary
method for analysts of all kinds of institutions, from the most particular
matters of legislative procedure to the broadest aspects of social order.
Viewing institutions as equilibria in some underlying, unalterable game
makes it possible for the first time to examine under a single model both
behavior within institutions and change of institutions.
A P PE N D I X : PROOFS OF LEMMA A N D T H EOREMS
Proof of Lemma: The proof uses induction to calculate the expected
discounted present value (DPV) of interactions with a given partner,
ignoring the current period; it is shown that this is equal to the sum over
t = 1 to infinity of ''/Y, where y is the payoff in each period and y is as
defined in the lemma. Thus y is the effective discount factor as required.
For notational convenience, let q = 1/(N - 1), the probability of meet­
ing a given partner on a given turn.
First, calculate the expected DPV of the payoff from the next single
interaction with the given player. This will be the sum from periods 1
through infinity following the present period of the probability of hav­
ing the next interaction with the partner in that period, times the dis­
counted payoff if that happens:
f (1 - q)'-lqf!y =y 1 }]? _\ = yy,
I;J
where y is as defined in the statement of the lemma.
Now, consider the expected DPV of the payoff from the (T + 1)-th
encounter with this partner after the current period, assuming that the
payoff from the T-th encounter is yry. The probability of the first
encounter taking place t periods after the present is (1 - q)'-1q; once
that happens in period t, the expected DPV (discounting from the
present period, t = 0) of the T-th encounter thereafter is flyry. Sum­
ming over t gives the expected DPV of the ( T + 1 )-th encounter from
the present:
L (1 - q )'-t q &yry = yryoq L (1 - q)'- 8
® ®
t t-1
t=l t=l
oq T
')' +ly .
yTy
8(1 - q)
= =
I -
84 Explaining Social Institutions
Thus the expected DPV of all future interaction with the given partner is
"' ., oq
2: I''Y = 2:
t=l t=l
[ 1 - 8(1 - q) ]/ y,
and substituting 1/(N - 1) for q gives the desired result. 0 Suppose that one player conformed to the equilibrium strategy on
Method for proofs of theorems. Ail the proofs below proceed by demon­
strating that there is no profitable one-period deviation from the equilib­
rium strategy; this includes deviations made after leaving the equilib­
rium path, since the equilibrium strategy specifies what is to be done in
those situations as well. To see why this is sufficient, consider the follow­
ing argument. When examining the incentives of a given player, the
strategies assumed for all of the other players, along with the structure
of the game, specify a dynamic programming problem for the given
player. This is an infinite-horizon problem with discounting, so if there
were some infinitely long sequence of departures from the specified
strategy that made the player better off, there would also be a finite
sequence of departures that improved payoff (since beyond some dis­
tant future point all further gains are minuscule once discounted back to
the present). Suppose there is no single-period departure that alone
makes the given player better off, however; then no longer-duration,
finite-length sequence of departures will do so either, for in the next-to­
last period of the sequence of departures, to depart for one more period
cannot be profitable. Thus, in order to show that there can be no profit­
able departure of any duration from the specified strategy by a single
player, it suffices to show that there is no profitable one-period depar­
ture.l9 Again, however, note that it is important to show that even if a
player has departed from the strategy, it must then be optimal to carry
out the appropriate off-equilibrium behavior specified by the strategy.
Proof of Theorem 1: Theorem 1 follows directly from the lemma,
along with standard results of the repeated prisoner's dilemma;20 it is
proved here for completeness, as well as to give a simple illustration of
the general technique of proof to be used subsequently. The standard
proof is given in terms of y, and the final result is obtained by applica­
tion of the lemma.
Suppose that both players cooperated on the previous iteration (or
that the game is in iteration 1). In such a situation, the DPV of all
present and future payoffs to a player from adhering to TFT by cooper­
ating on the current iteration, and then playing according to TFT there-
Rational Actors, Equilibrium, and Social Institutions 85
after (for a payoff of 1 in every period), can be written as 1 + 1' + "f-1
(1 - y) . The payoff from the one-period deviation of defecting now and
adhering to TFT in the future (including the consequent "restitution") is
ex - yf3 + "f-/(1 - y) . For cooperation to be the equilibrium move, then,
requires 1 + 1' ;::: a - yf3, or, 1' ;::: ( a - 1)1({3 + 1). Using the lemma to
substitute for 1' gives the first bound stated in theorem 1 .
the previous iteration, but that the other deviated. TFT then calls for
the conforming player to play D on the current iteration, which for any
y obviously gives a higher payoff than playing C since that D will not be
punished subsequently. Consider the decision of the deviant. To return
to the equilibrium path by playing C yields a payoff of -f3 + 1' + y2J
(1 - y) . The one-time deviation of playing D now and then returning to
TFT (that is, beginning "restitution" on the next iteration instead of the
present one) pays 0 - yf3 + y2/(1 - y). Thus for TFT to be in equilib­
rium requires - {3 + 1' ;::: - y{3, or 1' ;::: {3/(1 + {3). Applying the lemma
gives the second bound stated in theorem 1 .
These cases represent all of the possible situations in which a player
could consider a one-period deviation from TFT; in each case, the one­
period deviation would be unprofitable. Therefore the strategy profile
in which both players use TFT is an equilibrium. 0
Proof of Theorem 2: Phases 2 and 3 require only that the players
not engage in costly communication that will in any case be ignored
by other players using the assigned strategy. It remains only to show
that there are no profitable one-period deviations beginning in phases
4 or 5 .
In phase 4, if player i is in punishment status, playing C as pre­
scribed by TFT/MC yields an expected payoff (in discounted present
value from that point on) of
02
[ -{3 - (N - 2)c] + 8[1 - (N - 2)c] + 1 [1 - (N - 2)c],
8
_
while playing D and then returning to TFT/MC would yield a payoff of
52
0 + o[-{3 - (N - 2)c] + 1 _ 0 [1 - (N - 2)c].
Thus, in order for TFf/MC to be in equilibrium we must have the latter
no larger than the former, or
86 Explaining Social Institutions Rational Actors, Equilibrium, and Socia/ Institutions 87

- {3 - (N - 2)c + 8(1 - (N - 2)c] � o[-{3 - (N - 2)c]. which reduces to

By the second bound on o given in the theorem, this condition is satis­

fied. Note that such a o always exists since we assumed c < li(N - 2).
Consider next a player in cooperation status in phase 4. If the
player's partner is reported to be in punishment status, then obviously it Thus again, sufficiently large o < 1 satisfies the condition. Notice that
is optimal for the player to defect as prescribed. If the partner is re­ this right-hand side is smaller than that derived for cooperation status in
ported in cooperation status, playing C gives a payoff of [1 - (N - 2)c)/ phase 4, so the previous condition subsumes this one.
(1 - 8), which can be written as [1 - (N - 2)c] + 8[1 - (N - 2)c] + 52/ It remains to show that it is optimal in phase 5 to report to all of the
(1 - 8)[1 - (N - 2)c), while deviating to D gives remaining N - 2 players, rather than to just report to some of them. Let
K be the number of players to whom player i reports in phase 5, 0 ::::; K �
52 N - 2. Let Vob(K) represent the value of optimal continuation when
a - 8[{3 + (N - 2)c] +
1
_
0 [1 - (N - 2)c]. player i's status is a E {C,D}, i's current partner's status is b E {C,D}, i
reports to K other players in phase 5 of the current iteration and to N - 2
others thereafter, and all other players always report to all N - 2 other
The resulting necessary condition for equilibrium is then
players in phase 5. Then in the next iteration, the probability that player i
will meet one of the players to whom she reported is (K + 1)/ (N - 1)
[ 1 - (N - 2)c](1 + o) � a - 8[{3 + (N - 2)c],
(the K to whom she reported plus her current partner), while her proba­
bility of meeting a player to whom she did not report, thus having to
which reduces to make restitution, is (N - K - 2)/(N - 1). We can then write player i's
total expected payoff beginning with phase 5 in the current iteration as
a - 1 + (N - 2)c
o�
{3 + 1 [ K + 1 Vcc(N - 2) + N - K - 2 Vvc(N - 2) .
Since we assumed c < (2 - a + {3)/(N - 2), this can be satisfied by
Vob(K) = -Kc - 8
N
_ l
N
_ l
J
sufficiently large o < 1. The derivative of Vob(K) with respect to K is
Finally, turn to phase 5. Obviously a player in punishment status
will be content not to report, as prescribed. For a player in cooperation 8
status, reporting yields a payoff of -c + [Vcc(N - 2) - Vv c(N - 2)).
N _
1
0 82 1
-(N - 2)c + [ 1 - (N - 2)c] + - [ - (N - 2)c] , Clearly, Vcc(K) � Vvc(K) for all K sin ce in the former i's partner begins
1-8 1-8
-

by cooperating, while in the latter the partner begins by defecting.

Hence, as long as c is sufficiently small, the optimal K is as large as
while failing to report at all puts the player into punishment status and possible, that is, K N - 2. The bracketed term in the derivative
=

yields a payoff of reduces to 1 + {3, giving the third bound on c in the statement of the
theorem. 0
02 Note on the Proof of Theorem 2. There is no temptation for player i
0 + o [- {3 - (N - 2)c] + 0 [1 - (N - 2)c).
1 _

to falsely report that i's partner has played D in phase 5 since, because
the partner will not know that i lied, the partner will assume that she is
The resulting equilibrium condition is in cooperation status in the future and will not make restitution. If a
player's report could be made known to that player's partner, a tempta­
- 8{3 - o(N - 2)c � -(N - 2)c + cS[1 - (N - 2)c], tion to lie would be present- subgame perfection would require a
88 Explaining Social Institutions Rational Actors, Equilibrium, and Social Institutions 89

player falsely reported in punishment status to behave as though really N-2 [ -{3 - c + --
o N-2] 1
X= -- ( 1 - 2c) -- + -- oX,
in punishment status, so lying would pay off if the liar met the same N- 1 1 - 8 N- 1 N-1
partner on the very next iteration. Thus, if the report were assumed to
be known to the player's partner, the equilibrium behavior of theorem 2 for a total payoff of oX for defecting. (The complicated structure here is
could not be maintained without a modification of the equilibrium strat­ due to the fact that a deviant cannot make "restitution" until she is
egy to deter lying. This modification could be accomplished by the paired with a player other than the director; this occurs in any given
addition of a kind of tit-for-tat in truthful reporting, that is, a separate stage with probability (N - 2)/(N - 1).) Solving for X,

[ - {3 - c + ]
punishment scheme in which a player whose partner lies retaliates by
lying on the next meeting between the same two players in which both N-2 8 N-2
X= ( 2c) -
are in cooperation status. This would add a second condition on the N- 1 - o 1 -o 1 - N 1 '
discount factor, but the new condition would be less stringent than that
presently given in theorem 1 since, although the retaliation is heavily so the relevant equilibrium condition is

o N-2 o N-2 [ o N-2]

discounted, the reward from lying is itself discounted, accruing as it

1 - o 2c)--�
does only with probability 1/(N - 1 ) . Details of this proof are available
from the author on request. A similar modification would apply to
theorem 3 .
-{3 - c + -- (1 -
N-1 N-1-o -{3 - c +
1-o N-1
-- (1 - 2c) -- ,

which is true if and only if

Proof of Theorem 3 : The director has no incentive to violate any of
the strategy's prescriptions, so consider the incentives of players 2 o N-2
-{3 - c + -- (1 - 2c) -- � 0;
through N. In phase 2, obviously no such player in punishment status 1 - o N- 1
will wish to query. For a player in cooperation status, querying gives a
payoff of that is,

o N-2 {3 + c -
1 - 2c + 0 (1 - 2c) o�
1 N 1, N -2.
_ _

{3 + c + (1 - 2c) N 1 _

while failing to query yields The latter is always possible for sufficiently large o < 1 since c < V2.
If a player is in cooperation status in phase 4 and his partner is in
o N-2 punishment status, obviously there is no reason not to play D as pre­
-{3 - c - -- (1 - 2c) -- .
1 - o N-1 scribed. If both the player and the partner are in cooperation status, the
payoff to playing C is
The resulting condition for equilibrium is simply 1 - 2c � -{3 - c,
which is true since 0 < c < \12 and {3 > 0. o N-2
In phase 4 when the player is in punishment status, cooperating as 1 - c+ 0 (l - 2c)
1 N 1 ,
_
_

prescribed gives a payoff of

while the payoff from defecting is a + oX, where X is as defined above.
o N-2 Moving a to the left-hand side, the condition for equilibrium in coopera­
-{3 - c + -- (1 - 2c) --
1 - o N- 1 ' tion status in phase 4 becomes

o N-2 o(N-2) [ o N-
1-o N-1 N - 1 - 8 1-8 N-1 J
while defecting yields 0 in the present period and a payoff of X begin­ -a - c + 1
2
+ -- (1 - 2c) -- � -{3 - c + -- (1 - Zc) -- .
ning with the next period, where
90 Explaining Social Institutions
The resulting condition on o is thus
0> (N - 1)(a - 1 + c)
- (N - 3)({3 c + 1) + a + f3
- paid the phase 2 reporting fee, will be willing to pay even if that results
This is true for sufficiently large o < 1 , provided that the right-hand side
is less than 1 , which it is provided c < 1 - (a - {3)/2.
In phase 5, a player in punishment status obviously will not wish
to report. A player in cooperation status gains -c + (1 - 2c)o(N - 2)/
[(1 - o)(N - 1 )] by reporting, and 0 + oX by failing to report, where
X is as defined above. The resulting equilibrium condition is the same
as that derived above for punishment status in phase 4, except that -f3
is removed from the left-hand side. Thus the condition above is suffi­
cient to make the cooperation-status player report in phase 5 as well.
Finally, an argument similar to that explained in the note on the proof
of theorem 2 shows that a player has no incentive to report falsely in
phase 5 .
Proof of Theorem 4: To begin with, it is necessary to calculate the
optimal extortion payments for the director to demand when, off the
equilibrium path, a player is acquiescing to extortion. Once a player
acquiesces, he is forever extorted and forever acquiesces. On each itera­
tion, he pays the demand, is reported (falsely) to be in cooperation
status, plays D and does not report in phase 5. The only exception is
when he is paired with the director (probability li(N - 1)), or with the
same partner as was cheated on the previous turn, and who would
recognize the director's false report. The optimal payment to demand,
then, is one that leaves the victim just indifferent between continuing
this pattern and switching to permanent defection status, for a payment
of 0 in perpetuity. A payment of a - d - cpi(N - 3) accomplishes this
exactly: the extorted player's present value payoff for the remainder of
the game is
N - 3 <P - <P
N - 1 N - 3+N - 1
VP + oV
P
=
and solving yields simply VP 0. This extortion strategy yields the
= --
director a present value payoff of
N-3 1 1
Vd = N - 1 ( a - c) + N - 1 0 + N - 1 cp + 8Vd
Rational Actors, Equilibrium, and Social Institutions 91
or simply Vd (N - 3)(a - c)![(N - 1)(1 - o)]. Finally, this must be
=
compared to an alternative extortion startegy available to the director:
demand the largest possible payment that the player, having already
in the player subsequently entering permanent punishment phase. The
corresponding payment would be just a, the amount the player can get
from playing D on this iteration. For the director to eschew this larger
demand requires a ::::; Vd; solving this inequality for 8 yields the condi­
tion 8 � 1 - [(N - 3)/(N - I)][(a - c)/a]. Thus for sufficiently large o,
the strategy of ongoing extortion is superior for the corrupt director.
For the director, in phase 3, if a player has never before paid extor­
tion, then not extorting yields 2</J for the remainder of this period and 2</J
per period in future interactions with the player in question. An extortion
attempt yields 0 for the remainder of this turn and 2</J in the future,
so honesty is the best policy. If a player has paid extortion before,
not extorting yields 0 now and a - c - <Pf(N - 3) (that is, the
0 a - d - cp/(N - 3) extortion payment plus the <P in fees) per period
thereafter in periods when the player is not matched with the same part­
ner cheated on the previous turn (since according to this strategy the
director will subsequently resume extortion), while extorting yields a- d
now and a - c - <Pf(N - 3) per period in the future. Thus continued
extortion would be rational for the director. Finally, notice that to report
a nonpaying, cooperation-status player who refuses extortion as being in
punishment status would result in a loss of the phase 5 payments; hence
the director "forgives" a player who refuses to submit to extortion. (This
is why it was necessary to assume at the outset that the director would
carry out his one-period threat- even that would be irrational.)
Next consider the position of a player who has never paid extortion
and is now extorted for a payment B > 0. Submitting to the demand
would result in a future of constant, maximal extortion demands, so the
payoff to submitting would be a - B for the remainder of this period
and zero in every period thereafter. Refusing the demand and proceed­
ing according to AES would result in a payment of c to warn the partner
and no other payoff in the present period, and a resumption of coopera­
tion in the next period. The relevant equilibrium condition is then
8 N-2
-c + (1 - 2d) :::: a - B.
1 - 8 N-1
--
This must hold for any B > 0; hence the condition for deterring an
extortion attempt can be written by setting B 0 and solving for o,
=
which gives
92 Explaining Social Institutions
1 - 2d
5 "2::. N- 1 ·
1 - 2d + ( a - c) N 2
_
To show that it will be rational for such a player to bear the cost of
warning his partner, note that not warning results in the player entering
punishment status due to the corrupt director's report. To restore co­
operation (rational under TFT/CC given 8 sufficiently large), the ex­
torted player who does not warn will unilaterally play C, then report as
always in phase 5; cooperation is restored in the next iteration. Thus the
condition for warning one's partner is simply -c "2::. - f3 - d, which
always holds since f3 + cf> > 0.
Finally, suppose a player has paid extortion before, and now a - d
- cf>I(N 3) or less is demanded. Then not paying results in a payoff of
-
0 this turn, paying results in 0 or more, and all future periods will yield 0
in either case. If more than a d - cf>I(N - 3) is demanded the player
-
expects future demands of a. Obviously it is at least as good to be in
punishment status forever (payoff 0) as to continue this extortion pat­
tern. If the demand is no more than a, the player will pay it, play D, and
then enter permanent punishment status. Larger demands will result in
immediate refusal and permanent punishment status. 0 old players depart and new players enter the game.
NOTES
Earlier versions o f this chapter were presented a t the Public Choice Society
annual meetings, New Orleans, 20-22 March 1992, and at the Workshop on
New Institutional Theory at Cornell University, 8-10 November 1991. The au­
thor is grateful to Victor Nee, Andy Rutten, and other participants in the
workshop, as well as Jeffrey Banks, Jack Knight, Peter Stone, and David
Weimer, for encouragement, comments, and helpful conversations. Research
for this project was funded by the National Science Foundation through grants
SES-8908226 and BNS-9700864, the latter through the Center for Advanced
Study in the Behavioral Sciences, which generously supported the author as a
fellow during 1990-91 .
1 . As quoted by Hechter, Opp, and Wippler 1990, 1 .
2. A possible exception, close perhaps i n spirit to rational-actor theories, is
the approach of Robert Frank (1985, 1988), who attempts in several instances to
deriveunselfishpreferencesfromselfishonesviaevolutionaryarguments.Theorigin
of this idea lies in the explanation of "kinship altruism" given by Trivers (1974).
3. Alternatively, one might argue, along with March and Olsen (1989) and
Johnson (1990), that institutions represent the limits of individuals' abilities to
imagine alternative strategies, in effect specifying the strategy sets of the players
in a game. I balk at this explanation of institutions' ability to constrain behavior
Rational Actors, Equilibrium, and Social Institutions 93
because it assumes away the creativity that people often bring to bear in attempt­
ing to gain advantage under social constraints. My own game-theoretic ap­
proach is already weak enough in this respect.
4. As a halfway measure, one might assume that certain stable background
institutions define the underlying game, and then conduct an institutions-as­
equilibria analysis to study the formation of additional institutions. Schotter's
example of the institution of the week takes this form: trading institutions (e.g.,
the geographical location of a marketplace) are taken for granted, and the week
evolves as an equilibrium solution to a coordination problem concerning the
length of time between market days (Schetter 1981, 31-35)
5. This would rule out the often-cited legislative norm of reciprocity as an
institution if the reciprocity were solely a matter between pairs of legislators.
However, if a refusal to reciprocate with one partner were looked on by others
as an indication of lack of trustworthiness, then the reciprocity would in effect
be enforced by the whole group, and would be a real institution.
6. Normally, too, we consider a certain behavior pattern to be an institution
only if it persists over a long period of time. For present purposes, the require­
ment that an institution be an equilibrium pattern of behavior captures this
requirement. In a richer model than the one presented in this chapter, an
equilibrium should be considered an institution only if it persists despite the
occurrence of normal changes over time in the underlying game, such as when
7. "Social dilemma" is a term coined by Dawes, van de Kragt, and Orbell
(1990) to describe situations in which some form of group coordination or co­
operation can improve the welfare of all members of the group, but in which such
coordination or cooperation runs somehow counter to the short-run incentives or
abilities of the group's members. It includes prisoner's dilemma situations as well
as a multitude of other public good, externality, provision, and participation
problems.
8. The notation for G is identical to that used by Milgram, North, and
Weingast (1990) in order to facilitate comparisons.
9. This version of tit-for-tat was, to my kn owledge , first suggested by
Sugden (1986).
10. The equilibrium constructed in this section resembles the "Law Merchant
System Strategy" (LMSS) central to Milgrom, North, and Weingast (1990), in
its use of the pretrade (pre-PD) query and the central communicator. In the
LMSS, a player whose partner deviates from cooperative play registers a com­
plaint with the "judge," who assesses a fine against the deviant and reports the
deviant to be a noncooperator until the fine is paid. The present model dis­
penses with such fines, instead simply making the director the reporter of
whether a player is supposed to be punished, TFf style, in the play of the PD.
Further, the present analysis considers the cost of communication, while that of
Milgram, North, and Weingast does not. My main point, however, is to make a
direct comparison between this relatively highly structured equilibrium and the
less structured ones presented in sections 2 and 3.
 94 Explaining Social Institutions
1 1 . The reason for including the queries is so that the director can learn who
is playing whom (and thereby identify anyone who fails to report in phase 5)
without expending any search costs.
12. The reporting of extortion, it turns out, is possible under the same condi­
tion already required in theorem 5 in the text, using d.
13. The quote is from Krehbiel, Shepsle, and Weingast 1987, 936; see also
Shepsle and Weingast 1987, 85-86.
14. The argument that informal agreements in a legislature are likely to fall
prey to opportunism is first given in Weingast and Marshall 1988.
15. This effect, that the player selecting the coordination point has genuine
"power," plays an important general role in the nature of political leadership.
See Hardin 1990 and Calvert 1991.
16. Implicit in their theory is also the idea that changes in committee prefer­
ences can be used to predict changes in legislative outcomes, in a manner such
as that described by Weingast (1981).
17. Note that in both Sinclair and in the Baron-Ferejohn work, some institu­
tional features (e.g., basic parliamentary procedures) are taken as fixed so that
others (e.g., reciprocity norms or party arrangements) may be studied. Such a
"partial equilibrium" approach is likely to be the most useful in studying real­
life institutional features.
18. See, for example, the study by Samuelson (1991), which derives a corre­
spondence between "limit evolutionarily stable strategies," on the one hand,
and pure-strategy Nash equilibria among strategies not weakly dominated and
satisfying a property called "role-equivalence," on the other.
19. This is the strategy of proof adopted by Milgrom, North, and Weingast,
which they describe as an appeal to the optimality principle of dynamic program­
ming (1990, 8).
20. Such as Taylor 1976 or Axelrod 1981.
Viewpublicationstats
Models, Interpretations, and Theories:
Constructing Explanations of Institutional
Emergence and Change
Jack Knight
Rational-choice theories of institutional emergence and change take
many forms. They all share an initial premise that social actors pursue
some set of preferences in a rational way. This means that social actors
seek to achieve their most-preferred outcome in the least costly manner.
From this initial assumption, however, rational-choice theorists develop
quite different explanations of how social institutions emerge and
change. The main differences in the explanations of change lie in (1) the
different ways in which the theorists conceive and model the basic social
interactions that produce institutions and (2) the different features of the
social context that they invoke in order to resolve the strategic problems
inherent in these interactions (where the institutions are the product of
repeated efforts to resolve these problems).
The significance of these differences for explanations of institu­
tional change calls attention to the importance of some basic issues
related to the use of rational-choice models in the construction of social­
scientific explanations. Theory buil ding involves more than the develop­
ment of a model or the solving of a game. Explanations of social life
require the demonstration of the interrelationship between models and
the actual phenomenon to be explained. This requires that we give close
attention to the fact that models require interpretation. In the process of
interpretation, we assess the results derived from a model in light of
several factors, the most important of which for purposes of explanation
are (1) the initial assumptions used to develop the model, (2) the mecha­
nisms invoked to solve the model, and (3) the more general theoretical
framework of which the model and the resulting explanation are a part.
Furthermore, this process of interpretation involves a reconsideration of
both the assumptions and the mechanisms in light of the actual phenome­
non to be explained.
95